Merge pull request #46 from NVIDIA/p2p

Add alltoall perf test

src/common.cu

@@ -308,7 +308,7 @@ testResult_t CheckData(struct threadArgs* args, ncclDataType_t type, ncclRedOp_t
 #endif
   }
   double nranks = args->nProcs*args->nThreads*args->nGpus;
-  if (maxDelta > DeltaMaxValue(type)*(nranks - 1)) args->errors[0]++;
+  if (args->reportErrors && maxDelta > DeltaMaxValue(type)*(nranks - 1)) args->errors[0]++;
   *delta = maxDelta;
   return testSuccess;
 }
@@ -834,6 +834,8 @@ testResult_t run() {
     threads[t].args.bw=bw+t;
     threads[t].args.bw_count=bw_count+t;
 
+    threads[t].args.reportErrors = 1;
+
     threads[t].func = parallel_init ? threadInit : threadRunTests;
     if (t)
       TESTCHECK(threadLaunch(threads+t));